X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d1ecb069dfe24481e4a83f44cb5217a2b06746d7..bb59bff194111743b33cc36712410b5656329d3c:/bsd/kern/kern_descrip.c diff --git a/bsd/kern/kern_descrip.c b/bsd/kern/kern_descrip.c index d07556877..722fcfdff 100644 --- a/bsd/kern/kern_descrip.c +++ b/bsd/kern/kern_descrip.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2014 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -80,6 +80,7 @@ #include #include #include +#include #include #include #include @@ -92,8 +93,10 @@ #include #include #include -#include +#include #include +#include +#include #include @@ -101,6 +104,7 @@ #include #include #include +#include #include #include #include @@ -111,6 +115,12 @@ #include #include +#include + +#if CONFIG_PROTECT +#include +#endif +#include kern_return_t ipc_object_copyin(ipc_space_t, mach_port_name_t, mach_msg_type_name_t, ipc_port_t *); @@ -119,26 +129,19 @@ void ipc_port_release_send(ipc_port_t); struct psemnode; struct pshmnode; -int fdopen(dev_t dev, int mode, int type, proc_t p); -int finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval); +static int finishdup(proc_t p, + struct filedesc *fdp, int old, int new, int flags, int32_t *retval); int falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx, int locked); void fg_drop(struct fileproc * fp); void fg_free(struct fileglob *fg); void fg_ref(struct fileproc * fp); -#if CONFIG_EMBEDDED void fileport_releasefg(struct fileglob *fg); -#endif /* CONFIG_EMBEDDED */ /* flags for close_internal_locked */ #define FD_DUP2RESV 1 -static int close_internal_locked(struct proc *p, int fd, struct fileproc *fp, int flags); - -static int closef_finish(struct fileproc *fp, struct fileglob *fg, proc_t p, vfs_context_t ctx); /* We don't want these exported */ -__private_extern__ -int open1(vfs_context_t, struct nameidata *, int, struct vnode_attr *, int32_t *); __private_extern__ int unlink1(vfs_context_t, struct nameidata *, int); @@ -146,27 +149,27 @@ int unlink1(vfs_context_t, struct nameidata *, int); static void _fdrelse(struct proc * p, int fd); -extern void file_lock_init(void) __attribute__((section("__TEXT, initcode"))); -extern int kqueue_stat(struct fileproc *fp, void *ub, int isstat4, proc_t p); -#if SOCKETS -extern int soo_stat(struct socket *so, void *ub, int isstat64); -#endif /* SOCKETS */ +extern void file_lock_init(void); extern kauth_scope_t kauth_scope_fileop; -extern int cs_debug; +/* Conflict wait queue for when selects collide (opaque type) */ +extern struct wait_queue select_conflict_queue; #define f_flag f_fglob->fg_flag -#define f_type f_fglob->fg_type +#define f_type f_fglob->fg_ops->fo_type #define f_msgcount f_fglob->fg_msgcount #define f_cred f_fglob->fg_cred #define f_ops f_fglob->fg_ops #define f_offset f_fglob->fg_offset #define f_data f_fglob->fg_data +#define CHECK_ADD_OVERFLOW_INT64L(x, y) \ + (((((x) > 0) && ((y) > 0) && ((x) > LLONG_MAX - (y))) || \ + (((x) < 0) && ((y) < 0) && ((x) < LLONG_MIN - (y)))) \ + ? 1 : 0) /* * Descriptor management. */ -struct filelist filehead; /* head of list of open files */ struct fmsglist fmsghead; /* head of list of open files */ struct fmsglist fmsg_ithead; /* head of list of open files */ int nfiles; /* actual number of open files */ @@ -177,7 +180,64 @@ lck_grp_t * file_lck_grp; lck_attr_t * file_lck_attr; lck_mtx_t * uipc_lock; -lck_mtx_t * file_flist_lock; + + +/* + * check_file_seek_range + * + * Description: Checks if seek offsets are in the range of 0 to LLONG_MAX. + * + * Parameters: fl Flock structure. + * cur_file_offset Current offset in the file. + * + * Returns: 0 on Success. + * EOVERFLOW on overflow. + * EINVAL on offset less than zero. + */ + +static int +check_file_seek_range(struct flock *fl, off_t cur_file_offset) +{ + if (fl->l_whence == SEEK_CUR) { + /* Check if the start marker is beyond LLONG_MAX. */ + if (CHECK_ADD_OVERFLOW_INT64L(fl->l_start, cur_file_offset)) { + /* Check if start marker is negative */ + if (fl->l_start < 0) { + return EINVAL; + } + return EOVERFLOW; + } + /* Check if the start marker is negative. */ + if (fl->l_start + cur_file_offset < 0) { + return EINVAL; + } + /* Check if end marker is beyond LLONG_MAX. */ + if ((fl->l_len > 0) && (CHECK_ADD_OVERFLOW_INT64L(fl->l_start + + cur_file_offset, fl->l_len - 1))) { + return EOVERFLOW; + } + /* Check if the end marker is negative. */ + if ((fl->l_len <= 0) && (fl->l_start + cur_file_offset + + fl->l_len < 0)) { + return EINVAL; + } + } else if (fl->l_whence == SEEK_SET) { + /* Check if the start marker is negative. */ + if (fl->l_start < 0) { + return EINVAL; + } + /* Check if the end marker is beyond LLONG_MAX. */ + if ((fl->l_len > 0) && + CHECK_ADD_OVERFLOW_INT64L(fl->l_start, fl->l_len - 1)) { + return EOVERFLOW; + } + /* Check if the end marker is negative. */ + if ((fl->l_len < 0) && fl->l_start + fl->l_len < 0) { + return EINVAL; + } + } + return 0; +} /* @@ -203,7 +263,6 @@ file_lock_init(void) file_lck_attr = lck_attr_alloc_init(); uipc_lock = lck_mtx_alloc_init(file_lck_grp, file_lck_attr); - file_flist_lock = lck_mtx_alloc_init(file_lck_grp, file_lck_attr); } @@ -384,6 +443,7 @@ fd_rdwr( uio_t auio = NULL; char uio_buf[ UIO_SIZEOF(1) ]; struct vfs_context context = *(vfs_context_current()); + bool wrote_some = false; p = current_proc(); @@ -419,9 +479,11 @@ fd_rdwr( if ( !(io_flg & IO_APPEND)) flags = FOF_OFFSET; - if (rw == UIO_WRITE) + if (rw == UIO_WRITE) { + user_ssize_t orig_resid = uio_resid(auio); error = fo_write(fp, auio, flags, &context); - else + wrote_some = uio_resid(auio) < orig_resid; + } else error = fo_read(fp, auio, flags, &context); if (aresid) @@ -431,7 +493,7 @@ fd_rdwr( error = EIO; } out: - if (rw == UIO_WRITE && error == 0) + if (wrote_some) fp_drop_written(p, fd, fp); else fp_drop(p, fd, fp, 0); @@ -469,26 +531,31 @@ dup(proc_t p, struct dup_args *uap, int32_t *retval) proc_fdunlock(p); return(error); } + if (FP_ISGUARDED(fp, GUARD_DUP)) { + error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP); + (void) fp_drop(p, old, fp, 1); + proc_fdunlock(p); + return (error); + } if ( (error = fdalloc(p, 0, &new)) ) { fp_drop(p, old, fp, 1); proc_fdunlock(p); return (error); } - error = finishdup(p, fdp, old, new, retval); + error = finishdup(p, fdp, old, new, 0, retval); fp_drop(p, old, fp, 1); proc_fdunlock(p); return (error); } - /* * dup2 * * Description: Duplicate a file descriptor to a particular value. * * Parameters: p Process performing the dup - * uap->fd The fd to dup + * uap->from The fd to dup * uap->to The fd to dup it to * retval Pointer to the call return area * @@ -513,6 +580,12 @@ startover: proc_fdunlock(p); return(error); } + if (FP_ISGUARDED(fp, GUARD_DUP)) { + error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP); + (void) fp_drop(p, old, fp, 1); + proc_fdunlock(p); + return (error); + } if (new < 0 || (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || new >= maxfiles) { @@ -547,8 +620,16 @@ closeit: goto startover; } - if ((fdp->fd_ofiles[new] != NULL) && ((error = fp_lookup(p, new, &nfp, 1)) == 0)) { + if ((fdp->fd_ofiles[new] != NULL) && + ((error = fp_lookup(p, new, &nfp, 1)) == 0)) { fp_drop(p, old, fp, 1); + if (FP_ISGUARDED(nfp, GUARD_CLOSE)) { + error = fp_guard_exception(p, + new, nfp, kGUARD_EXC_CLOSE); + (void) fp_drop(p, new, nfp, 1); + proc_fdunlock(p); + return (error); + } (void)close_internal_locked(p, new, nfp, FD_DUP2RESV); #if DIAGNOSTIC proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED); @@ -558,7 +639,7 @@ closeit: } else { #if DIAGNOSTIC if (fdp->fd_ofiles[new] != NULL) - panic("dup2: unable to get ref on a fileproc %d\n", new); + panic("dup2: no ref on fileproc %d", new); #endif procfdtbl_reservefd(p, new); } @@ -570,11 +651,11 @@ closeit: } #if DIAGNOSTIC if (fdp->fd_ofiles[new] != 0) - panic("dup2-1: overwriting fd_ofiles with new %d\n", new); + panic("dup2: overwriting fd_ofiles with new %d", new); if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) - panic("dup2-1: unreserved fileflags with new %d\n", new); + panic("dup2: unreserved fileflags with new %d", new); #endif - error = finishdup(p, fdp, old, new, retval); + error = finishdup(p, fdp, old, new, 0, retval); fp_drop(p, old, fp, 1); proc_fdunlock(p); @@ -638,6 +719,7 @@ fcntl(proc_t p, struct fcntl_args *uap, int32_t *retval) * copyin:EFAULT * vnode_getwithref:??? * VNOP_ADVLOCK:??? + * msleep:ETIMEDOUT * [F_GETLK] * EBADF * EOVERFLOW @@ -674,11 +756,12 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */ int i, tmp, error, error2, flg = F_POSIX; struct flock fl; + struct flocktimeout fltimeout; + struct timespec *timeout = NULL; struct vfs_context context; off_t offset; int newmin; daddr64_t lbn, bn; - int devBlockSize = 0; unsigned int fflag; user_addr_t argp; boolean_t is64bit; @@ -723,6 +806,11 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) switch (uap->cmd) { case F_DUPFD: + case F_DUPFD_CLOEXEC: + if (FP_ISGUARDED(fp, GUARD_DUP)) { + error = fp_guard_exception(p, fd, fp, kGUARD_EXC_DUP); + goto out; + } newmin = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */ AUDIT_ARG(value32, newmin); if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur || @@ -732,18 +820,27 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) } if ( (error = fdalloc(p, newmin, &i)) ) goto out; - error = finishdup(p, fdp, fd, i, retval); + error = finishdup(p, fdp, fd, i, + uap->cmd == F_DUPFD_CLOEXEC ? UF_EXCLOSE : 0, retval); goto out; case F_GETFD: - *retval = (*pop & UF_EXCLOSE)? 1 : 0; + *retval = (*pop & UF_EXCLOSE)? FD_CLOEXEC : 0; error = 0; goto out; case F_SETFD: AUDIT_ARG(value32, uap->arg); - *pop = (*pop &~ UF_EXCLOSE) | - (uap->arg & 1)? UF_EXCLOSE : 0; + if (uap->arg & FD_CLOEXEC) + *pop |= UF_EXCLOSE; + else { + if (FILEPROC_TYPE(fp) == FTYPE_GUARDED) { + error = fp_guard_exception(p, + fd, fp, kGUARD_EXC_NOCLOEXEC); + goto out; + } + *pop &= ~UF_EXCLOSE; + } error = 0; goto out; @@ -789,7 +886,7 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) goto out; } if (fp->f_type == DTYPE_PIPE) { - error = fo_ioctl(fp, (int)TIOCSPGRP, (caddr_t)&tmp, &context); + error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context); goto out; } @@ -807,6 +904,45 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) error = fo_ioctl(fp, (int)TIOCSPGRP, (caddr_t)&tmp, &context); goto out; + case F_SETNOSIGPIPE: + tmp = CAST_DOWN_EXPLICIT(int, uap->arg); + if (fp->f_type == DTYPE_SOCKET) { +#if SOCKETS + error = sock_setsockopt((struct socket *)fp->f_data, + SOL_SOCKET, SO_NOSIGPIPE, &tmp, sizeof (tmp)); +#else + error = EINVAL; +#endif + } else { + struct fileglob *fg = fp->f_fglob; + + lck_mtx_lock_spin(&fg->fg_lock); + if (tmp) + fg->fg_lflags |= FG_NOSIGPIPE; + else + fg->fg_lflags &= FG_NOSIGPIPE; + lck_mtx_unlock(&fg->fg_lock); + error = 0; + } + goto out; + + case F_GETNOSIGPIPE: + if (fp->f_type == DTYPE_SOCKET) { +#if SOCKETS + int retsize = sizeof (*retval); + error = sock_getsockopt((struct socket *)fp->f_data, + SOL_SOCKET, SO_NOSIGPIPE, retval, &retsize); +#else + error = EINVAL; +#endif + } else { + *retval = (fp->f_fglob->fg_lflags & FG_NOSIGPIPE) ? + 1 : 0; + error = 0; + } + goto out; + + case F_SETLKWTIMEOUT: case F_SETLKW: flg |= F_WAIT; /* Fall into F_SETLK */ @@ -823,14 +959,25 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) proc_fdunlock(p); /* Copy in the lock structure */ - error = copyin(argp, (caddr_t)&fl, sizeof(fl)); - if (error) { - goto outdrop; + if (uap->cmd == F_SETLKWTIMEOUT) { + error = copyin(argp, (caddr_t) &fltimeout, sizeof(fltimeout)); + if (error) { + goto outdrop; + } + fl = fltimeout.fl; + timeout = &fltimeout.timeout; + } else { + error = copyin(argp, (caddr_t)&fl, sizeof(fl)); + if (error) { + goto outdrop; + } } - if ((fl.l_whence == SEEK_CUR) && (fl.l_start + offset < fl.l_start)) { - error = EOVERFLOW; - goto outdrop; + /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */ + /* and ending byte for EOVERFLOW in SEEK_SET */ + error = check_file_seek_range(&fl, offset); + if (error) { + goto outdrop; } if ( (error = vnode_getwithref(vp)) ) { @@ -857,7 +1004,7 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) } // XXX UInt32 unsafe for LP64 kernel OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag); - error = VNOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg, &context); + error = VNOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg, &context, timeout); (void)vnode_put(vp); goto outdrop; @@ -869,13 +1016,13 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) } // XXX UInt32 unsafe for LP64 kernel OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag); - error = VNOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg, &context); + error = VNOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg, &context, timeout); (void)vnode_put(vp); goto outdrop; case F_UNLCK: error = VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl, - F_POSIX, &context); + F_POSIX, &context, timeout); (void)vnode_put(vp); goto outdrop; @@ -902,13 +1049,8 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */ /* and ending byte for EOVERFLOW in SEEK_SET */ - if (((fl.l_whence == SEEK_CUR) && - ((fl.l_start + offset < fl.l_start) || - ((fl.l_len > 0) && (fl.l_start+offset + fl.l_len - 1 < fl.l_start+offset)))) || - ((fl.l_whence == SEEK_SET) && (fl.l_len > 0) && (fl.l_start + fl.l_len - 1 < fl.l_start))) - { - /* lf_advlock doesn't check start/end for F_GETLK if file has no locks */ - error = EOVERFLOW; + error = check_file_seek_range(&fl, offset); + if (error) { goto outdrop; } @@ -943,10 +1085,10 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) #if CONFIG_MACF error = mac_file_check_lock(proc_ucred(p), fp->f_fglob, - F_GETLK, &fl); + uap->cmd, &fl); if (error == 0) #endif - error = VNOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX, &context); + error = VNOP_ADVLOCK(vp, (caddr_t)p, uap->cmd, &fl, F_POSIX, &context, NULL); (void)vnode_put(vp); @@ -1071,7 +1213,7 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) * without zero filling the data is a security hole * root would have access anyway so we'll allow it */ - if (!is_suser()) { + if (!kauth_cred_issuser(kauth_cred_get())) { error = EACCES; } else { /* @@ -1108,6 +1250,30 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) goto out; + case F_NODIRECT: + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + if (uap->arg) + fp->f_fglob->fg_flag |= FNODIRECT; + else + fp->f_fglob->fg_flag &= ~FNODIRECT; + + goto out; + + case F_SINGLE_WRITER: + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + if (uap->arg) + fp->f_fglob->fg_flag |= FSINGLE_WRITER; + else + fp->f_fglob->fg_flag &= ~FSINGLE_WRITER; + + goto out; + case F_GLOBAL_NOCACHE: if (fp->f_type != DTYPE_VNODE) { error = EBADF; @@ -1170,60 +1336,39 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) goto outdrop; } - case F_READBOOTSTRAP: - case F_WRITEBOOTSTRAP: { - user32_fbootstraptransfer_t user32_fbt_struct; - user_fbootstraptransfer_t user_fbt_struct; - int sizeof_struct; - caddr_t boot_structp; + case F_FLUSH_DATA: - if (fp->f_type != DTYPE_VNODE) { - error = EBADF; - goto out; - } - vp = (struct vnode *)fp->f_data; - proc_fdunlock(p); + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); - if (IS_64BIT_PROCESS(p)) { - sizeof_struct = sizeof(user_fbt_struct); - boot_structp = (caddr_t) &user_fbt_struct; - } - else { - sizeof_struct = sizeof(user32_fbt_struct); - boot_structp = (caddr_t) &user32_fbt_struct; - } - error = copyin(argp, boot_structp, sizeof_struct); - if (error) - goto outdrop; - if ( (error = vnode_getwithref(vp)) ) { - goto outdrop; - } - if (uap->cmd == F_WRITEBOOTSTRAP) { - /* - * Make sure that we are root. Updating the - * bootstrap on a disk could be a security hole - */ - if (!is_suser()) { - (void)vnode_put(vp); - error = EACCES; - goto outdrop; - } - } - if (strncmp(vnode_mount(vp)->mnt_vfsstat.f_fstypename, "hfs", - sizeof(vnode_mount(vp)->mnt_vfsstat.f_fstypename)) != 0) { - error = EINVAL; - } else { - /* - * call vnop_ioctl to handle the I/O - */ - error = VNOP_IOCTL(vp, uap->cmd, boot_structp, 0, &context); - } - (void)vnode_put(vp); - goto outdrop; - } - case F_LOG2PHYS: { + if ( (error = vnode_getwithref(vp)) == 0 ) { + error = cluster_push(vp, 0); + + (void)vnode_put(vp); + } + goto outdrop; + + case F_LOG2PHYS: + case F_LOG2PHYS_EXT: { struct log2phys l2p_struct; /* structure for allocate command */ + int devBlockSize; + off_t file_offset = 0; + size_t a_size = 0; + size_t run = 0; + + if (uap->cmd == F_LOG2PHYS_EXT) { + error = copyin(argp, (caddr_t)&l2p_struct, sizeof(l2p_struct)); + if (error) + goto out; + file_offset = l2p_struct.l2p_devoffset; + } else { + file_offset = fp->f_offset; + } if (fp->f_type != DTYPE_VNODE) { error = EBADF; goto out; @@ -1233,7 +1378,7 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) if ( (error = vnode_getwithref(vp)) ) { goto outdrop; } - error = VNOP_OFFTOBLK(vp, fp->f_offset, &lbn); + error = VNOP_OFFTOBLK(vp, file_offset, &lbn); if (error) { (void)vnode_put(vp); goto outdrop; @@ -1244,16 +1389,52 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) goto outdrop; } devBlockSize = vfs_devblocksize(vnode_mount(vp)); - - error = VNOP_BLOCKMAP(vp, offset, devBlockSize, &bn, NULL, NULL, 0, &context); + if (uap->cmd == F_LOG2PHYS_EXT) { +#if defined(__LP64__) + a_size = l2p_struct.l2p_contigbytes; +#else + if ((l2p_struct.l2p_contigbytes > SIZE_MAX) || (l2p_struct.l2p_contigbytes < 0)) { + /* size_t is 32-bit on a 32-bit kernel, therefore + * assigning l2p_contigbytes to a_size may have + * caused integer overflow. We, therefore, return + * an error here instead of calculating incorrect + * value. + */ + printf ("fcntl: F_LOG2PHYS_EXT: l2p_contigbytes=%lld will overflow, returning error\n", l2p_struct.l2p_contigbytes); + error = EFBIG; + goto outdrop; + } else { + a_size = l2p_struct.l2p_contigbytes; + } +#endif + } else { + a_size = devBlockSize; + } + + error = VNOP_BLOCKMAP(vp, offset, a_size, &bn, &run, NULL, 0, &context); (void)vnode_put(vp); if (!error) { l2p_struct.l2p_flags = 0; /* for now */ - l2p_struct.l2p_contigbytes = 0; /* for now */ - l2p_struct.l2p_devoffset = bn * devBlockSize; - l2p_struct.l2p_devoffset += fp->f_offset - offset; + if (uap->cmd == F_LOG2PHYS_EXT) { + l2p_struct.l2p_contigbytes = run - (file_offset - offset); + } else { + l2p_struct.l2p_contigbytes = 0; /* for now */ + } + + /* + * The block number being -1 suggests that the file offset is not backed + * by any real blocks on-disk. As a result, just let it be passed back up wholesale. + */ + if (bn == -1) { + /* Don't multiply it by the block size */ + l2p_struct.l2p_devoffset = bn; + } + else { + l2p_struct.l2p_devoffset = bn * devBlockSize; + l2p_struct.l2p_devoffset += file_offset - offset; + } error = copyout((caddr_t)&l2p_struct, argp, sizeof(l2p_struct)); } goto outdrop; @@ -1384,11 +1565,12 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) VATTR_SET(&va, va_mode, cmode & ACCESSPERMS); /* Start the lookup relative to the file descriptor's vnode. */ - NDINIT(&nd, LOOKUP, USEDVP | FOLLOW | AUDITVNPATH1, UIO_USERSPACE, + NDINIT(&nd, LOOKUP, OP_OPEN, USEDVP | FOLLOW | AUDITVNPATH1, UIO_USERSPACE, fopen.o_pathname, &context); nd.ni_dvp = vp; - error = open1(&context, &nd, fopen.o_flags, &va, retval); + error = open1(&context, &nd, fopen.o_flags, &va, + fileproc_alloc_init, NULL, retval); vnode_put(vp); break; @@ -1429,7 +1611,8 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) } /* Start the lookup relative to the file descriptor's vnode. */ - NDINIT(&nd, DELETE, USEDVP | AUDITVNPATH1, UIO_USERSPACE, pathname, &context); + NDINIT(&nd, DELETE, OP_UNLINK, USEDVP | AUDITVNPATH1, UIO_USERSPACE, + pathname, &context); nd.ni_dvp = vp; error = unlink1(&context, &nd, 0); @@ -1475,15 +1658,19 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) if(ubc_cs_blob_get(vp, CPU_TYPE_ANY, fs.fs_file_start)) { - /* - if(cs_debug) - printf("CODE SIGNING: resident blob offered for: %s\n", vp->v_name); - */ vnode_put(vp); goto outdrop; } - -#define CS_MAX_BLOB_SIZE (1ULL * 1024 * 1024) /* XXX ? */ +/* + * An arbitrary limit, to prevent someone from mapping in a 20GB blob. This should cover + * our use cases for the immediate future, but note that at the time of this commit, some + * platforms are nearing 2MB blob sizes (with a prior soft limit of 2.5MB). + * + * We should consider how we can manage this more effectively; the above means that some + * platforms are using megabytes of memory for signing data; it merely hasn't crossed the + * threshold considered ridiculous at the time of this change. + */ +#define CS_MAX_BLOB_SIZE (10ULL * 1024ULL * 1024ULL) if (fs.fs_blob_size > CS_MAX_BLOB_SIZE) { error = E2BIG; vnode_put(vp); @@ -1533,86 +1720,691 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) kernel_blob_size); } else { /* ubc_blob_add() has consumed "kernel_blob_addr" */ +#if CHECK_CS_VALIDATION_BITMAP + ubc_cs_validation_bitmap_allocate( vp ); +#endif } (void) vnode_put(vp); break; } + case F_FINDSIGS: { +#ifdef SECURE_KERNEL + error = ENOTSUP; +#else /* !SECURE_KERNEL */ + off_t offsetMacho; - case F_MARKDEPENDENCY: { - struct vnode *root_vp; - struct vnode_attr va; + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); + error = vnode_getwithref(vp); + if (error) + goto outdrop; + + error = copyin(argp, &offsetMacho, sizeof(offsetMacho)); + if (error) { + (void)vnode_put(vp); + goto outdrop; + } + +#if CONFIG_MACF + error = mac_vnode_find_sigs(p, vp, offsetMacho); +#else + error = EPERM; +#endif + if (error) { + (void)vnode_put(vp); + goto outdrop; + } +#endif /* SECURE_KERNEL */ + break; + } +#if CONFIG_PROTECT + case F_GETPROTECTIONCLASS: { + int class = 0; + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + + proc_fdunlock(p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + error = cp_vnode_getclass (vp, &class); + if (error == 0) { + *retval = class; + } + + vnode_put(vp); + break; + } + + case F_SETPROTECTIONCLASS: { + /* tmp must be a valid PROTECTION_CLASS_* */ + tmp = CAST_DOWN_EXPLICIT(uint32_t, uap->arg); + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + + proc_fdunlock(p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + /* Only go forward if you have write access */ vfs_context_t ctx = vfs_context_current(); - kauth_cred_t cred; + if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + error = cp_vnode_setclass (vp, tmp); + vnode_put(vp); + break; + } + + case F_TRANSCODEKEY: { + + char *backup_keyp = NULL; + unsigned backup_key_len = CP_MAX_WRAPPEDKEYSIZE; - if ((current_proc()->p_flag & P_DEPENDENCY_CAPABLE) == 0) { - error = EPERM; - goto out; + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; } + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + MALLOC(backup_keyp, char *, backup_key_len, M_TEMP, M_WAITOK); + if (backup_keyp == NULL) { + error = ENOMEM; + goto outdrop; + } + + error = cp_vnode_transcode (vp, backup_keyp, &backup_key_len); + vnode_put(vp); + + if (error == 0) { + error = copyout((caddr_t)backup_keyp, argp, backup_key_len); + *retval = backup_key_len; + } + + FREE(backup_keyp, M_TEMP); + + break; + } + + case F_GETPROTECTIONLEVEL: { + uint32_t cp_version = 0; + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + vp = (struct vnode*) fp->f_data; + proc_fdunlock (p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + /* + * if cp_get_major_vers fails, error will be set to proper errno + * and cp_version will still be 0. + */ + + error = cp_get_root_major_vers (vp, &cp_version); + *retval = cp_version; + + vnode_put (vp); + break; + } + + case F_GETDEFAULTPROTLEVEL: { + uint32_t cp_default = 0; + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + vp = (struct vnode*) fp->f_data; + proc_fdunlock (p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + /* + * if cp_get_major_vers fails, error will be set to proper errno + * and cp_version will still be 0. + */ + + error = cp_get_default_level(vp, &cp_default); + *retval = cp_default; + + vnode_put (vp); + break; + } + + +#endif /* CONFIG_PROTECT */ + + case F_MOVEDATAEXTENTS: { + struct fileproc *fp2 = NULL; + struct vnode *src_vp = NULLVP; + struct vnode *dst_vp = NULLVP; + /* We need to grab the 2nd FD out of the argments before moving on. */ + int fd2 = CAST_DOWN_EXPLICIT(int32_t, uap->arg); + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + /* For now, special case HFS+ only, since this is SPI. */ + src_vp = (struct vnode *)fp->f_data; + if (src_vp->v_tag != VT_HFS) { + error = EINVAL; + goto out; + } + + /* + * Get the references before we start acquiring iocounts on the vnodes, + * while we still hold the proc fd lock + */ + if ( (error = fp_lookup(p, fd2, &fp2, 1)) ) { + error = EBADF; + goto out; + } + if (fp2->f_type != DTYPE_VNODE) { + fp_drop(p, fd2, fp2, 1); + error = EBADF; + goto out; + } + dst_vp = (struct vnode *)fp2->f_data; + if (dst_vp->v_tag != VT_HFS) { + fp_drop(p, fd2, fp2, 1); + error = EINVAL; + goto out; + } + +#if CONFIG_MACF + /* Re-do MAC checks against the new FD, pass in a fake argument */ + error = mac_file_check_fcntl(proc_ucred(p), fp2->f_fglob, uap->cmd, 0); + if (error) { + fp_drop(p, fd2, fp2, 1); + goto out; + } +#endif + /* Audit the 2nd FD */ + AUDIT_ARG(fd, fd2); + + proc_fdunlock(p); + + if (vnode_getwithref(src_vp)) { + fp_drop(p, fd2, fp2, 0); + error = ENOENT; + goto outdrop; + } + if (vnode_getwithref(dst_vp)) { + vnode_put (src_vp); + fp_drop(p, fd2, fp2, 0); + error = ENOENT; + goto outdrop; + } + + /* + * Basic asserts; validate they are not the same and that + * both live on the same filesystem. + */ + if (dst_vp == src_vp) { + vnode_put (src_vp); + vnode_put (dst_vp); + fp_drop (p, fd2, fp2, 0); + error = EINVAL; + goto outdrop; + } + + if (dst_vp->v_mount != src_vp->v_mount) { + vnode_put (src_vp); + vnode_put (dst_vp); + fp_drop (p, fd2, fp2, 0); + error = EXDEV; + goto outdrop; + } + + /* Now we have a legit pair of FDs. Go to work */ + + /* Now check for write access to the target files */ + if(vnode_authorize(src_vp, NULLVP, + (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) { + vnode_put(src_vp); + vnode_put(dst_vp); + fp_drop(p, fd2, fp2, 0); + error = EBADF; + goto outdrop; + } + + if(vnode_authorize(dst_vp, NULLVP, + (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) { + vnode_put(src_vp); + vnode_put(dst_vp); + fp_drop(p, fd2, fp2, 0); + error = EBADF; + goto outdrop; + } + + /* Verify that both vps point to files and not directories */ + if ( !vnode_isreg(src_vp) || !vnode_isreg(dst_vp)) { + error = EINVAL; + vnode_put (src_vp); + vnode_put (dst_vp); + fp_drop (p, fd2, fp2, 0); + goto outdrop; + } + + /* + * The exchangedata syscall handler passes in 0 for the flags to VNOP_EXCHANGE. + * We'll pass in our special bit indicating that the new behavior is expected + */ + + error = VNOP_EXCHANGE(src_vp, dst_vp, FSOPT_EXCHANGE_DATA_ONLY, &context); + + vnode_put (src_vp); + vnode_put (dst_vp); + fp_drop(p, fd2, fp2, 0); + break; + } + + /* + * SPI for making a file compressed. + */ + case F_MAKECOMPRESSED: { + uint32_t gcounter = CAST_DOWN_EXPLICIT(uint32_t, uap->arg); + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + vp = (struct vnode*) fp->f_data; + proc_fdunlock (p); + + /* get the vnode */ + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + /* Is it a file? */ + if ((vnode_isreg(vp) == 0) && (vnode_islnk(vp) == 0)) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + + /* invoke ioctl to pass off to FS */ + /* Only go forward if you have write access */ + vfs_context_t ctx = vfs_context_current(); + if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + + error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)&gcounter, 0, &context); + + vnode_put (vp); + break; + } + + /* + * SPI (private) for indicating to a filesystem that subsequent writes to + * the open FD will written to the Fastflow. + */ + case F_SET_GREEDY_MODE: + /* intentionally drop through to the same handler as F_SETSTATIC. + * both fcntls should pass the argument and their selector into VNOP_IOCTL. + */ + + /* + * SPI (private) for indicating to a filesystem that subsequent writes to + * the open FD will represent static content. + */ + case F_SETSTATICCONTENT: { + caddr_t ioctl_arg = NULL; + + if (uap->arg) { + ioctl_arg = (caddr_t) 1; + } + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); + + error = vnode_getwithref(vp); + if (error) { + error = ENOENT; + goto outdrop; + } + + /* Only go forward if you have write access */ + vfs_context_t ctx = vfs_context_current(); + if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + + error = VNOP_IOCTL(vp, uap->cmd, ioctl_arg, 0, &context); + (void)vnode_put(vp); + + break; + } + + /* + * SPI (private) for indicating to the lower level storage driver that the + * subsequent writes should be of a particular IO type (burst, greedy, static), + * or other flavors that may be necessary. + */ + case F_SETIOTYPE: { + caddr_t param_ptr; + uint32_t param; + + if (uap->arg) { + /* extract 32 bits of flags from userland */ + param_ptr = (caddr_t) uap->arg; + param = (uint32_t) param_ptr; + } + else { + /* If no argument is specified, error out */ + error = EINVAL; + goto out; + } + + /* + * Validate the different types of flags that can be specified: + * all of them are mutually exclusive for now. + */ + switch (param) { + case F_IOTYPE_ISOCHRONOUS: + break; + + default: + error = EINVAL; + goto out; + } + + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); + + error = vnode_getwithref(vp); + if (error) { + error = ENOENT; + goto outdrop; + } + + /* Only go forward if you have write access */ + vfs_context_t ctx = vfs_context_current(); + if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + + error = VNOP_IOCTL(vp, uap->cmd, param_ptr, 0, &context); + (void)vnode_put(vp); + + break; + } + + + /* + * Extract the CodeDirectory of the vnode associated with + * the file descriptor and copy it back to user space + */ + case F_GETCODEDIR: { + struct user_fcodeblobs args; + + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + vp = (struct vnode *)fp->f_data; + proc_fdunlock(p); + + if ((fp->f_flag & FREAD) == 0) { + error = EBADF; + goto outdrop; + } + + if (IS_64BIT_PROCESS(p)) { + struct user64_fcodeblobs args64; + + error = copyin(argp, &args64, sizeof(args64)); + if (error) + goto outdrop; + + args.f_cd_hash = args64.f_cd_hash; + args.f_hash_size = args64.f_hash_size; + args.f_cd_buffer = args64.f_cd_buffer; + args.f_cd_size = args64.f_cd_size; + args.f_out_size = args64.f_out_size; + args.f_arch = args64.f_arch; + } else { + struct user32_fcodeblobs args32; + + error = copyin(argp, &args32, sizeof(args32)); + if (error) + goto outdrop; + + args.f_cd_hash = CAST_USER_ADDR_T(args32.f_cd_hash); + args.f_hash_size = args32.f_hash_size; + args.f_cd_buffer = CAST_USER_ADDR_T(args32.f_cd_buffer); + args.f_cd_size = args32.f_cd_size; + args.f_out_size = CAST_USER_ADDR_T(args32.f_out_size); + args.f_arch = args32.f_arch; + } + + if (vp->v_ubcinfo == NULL) { + error = EINVAL; + goto outdrop; + } + + struct cs_blob *t_blob = vp->v_ubcinfo->cs_blobs; + + /* + * This call fails if there is no cs_blob corresponding to the + * vnode, or if there are multiple cs_blobs present, and the caller + * did not specify which cpu_type they want the cs_blob for + */ + if (t_blob == NULL) { + error = ENOENT; /* there is no codesigning blob for this process */ + goto outdrop; + } else if (args.f_arch == 0 && t_blob->csb_next != NULL) { + error = ENOENT; /* too many architectures and none specified */ + goto outdrop; + } + + /* If the user specified an architecture, find the right blob */ + if (args.f_arch != 0) { + while (t_blob) { + if (t_blob->csb_cpu_type == args.f_arch) + break; + t_blob = t_blob->csb_next; + } + /* The cpu_type the user requested could not be found */ + if (t_blob == NULL) { + error = ENOENT; + goto outdrop; + } + } + + const CS_SuperBlob *super_blob = (void *)t_blob->csb_mem_kaddr; + const CS_CodeDirectory *cd = findCodeDirectory(super_blob, + (char *) super_blob, + (char *) super_blob + t_blob->csb_mem_size); + if (cd == NULL) { + error = ENOENT; + goto outdrop; + } + + uint64_t buffer_size = ntohl(cd->length); + + if (buffer_size > UINT_MAX) { + error = ERANGE; + goto outdrop; + } + + error = copyout(&buffer_size, args.f_out_size, sizeof(unsigned int)); + if (error) + goto outdrop; + + if (sizeof(t_blob->csb_sha1) > args.f_hash_size || + buffer_size > args.f_cd_size) { + error = ERANGE; + goto outdrop; + } + + error = copyout(t_blob->csb_sha1, args.f_cd_hash, sizeof(t_blob->csb_sha1)); + if (error) + goto outdrop; + error = copyout(cd, args.f_cd_buffer, buffer_size); + if (error) + goto outdrop; + + break; + } + + /* + * Set the vnode pointed to by 'fd' + * and tag it as the (potentially future) backing store + * for another filesystem + */ + case F_SETBACKINGSTORE: { + if (fp->f_type != DTYPE_VNODE) { + error = EBADF; + goto out; + } + + vp = (struct vnode *)fp->f_data; + + if (vp->v_tag != VT_HFS) { + error = EINVAL; + goto out; + } + proc_fdunlock(p); + + if (vnode_getwithref(vp)) { + error = ENOENT; + goto outdrop; + } + + /* only proceed if you have write access */ + vfs_context_t ctx = vfs_context_current(); + if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) { + vnode_put(vp); + error = EBADF; + goto outdrop; + } + + + /* If arg != 0, set, otherwise unset */ + if (uap->arg) { + error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)1, 0, &context); + } + else { + error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)NULL, 0, &context); + } + + vnode_put(vp); + break; + } + + /* + * like F_GETPATH, but special semantics for + * the mobile time machine handler. + */ + case F_GETPATH_MTMINFO: { + char *pathbufp; + int pathlen; + if (fp->f_type != DTYPE_VNODE) { error = EBADF; goto out; } - vp = (struct vnode *)fp->f_data; proc_fdunlock(p); - if (vnode_getwithref(vp)) { - error = ENOENT; + pathlen = MAXPATHLEN; + MALLOC(pathbufp, char *, pathlen, M_TEMP, M_WAITOK); + if (pathbufp == NULL) { + error = ENOMEM; goto outdrop; } + if ( (error = vnode_getwithref(vp)) == 0 ) { + int backingstore = 0; + + /* Check for error from vn_getpath before moving on */ + if ((error = vn_getpath(vp, pathbufp, &pathlen)) == 0) { + if (vp->v_tag == VT_HFS) { + error = VNOP_IOCTL (vp, uap->cmd, (caddr_t) &backingstore, 0, &context); + } + (void)vnode_put(vp); - // the passed in vnode must be the root dir of the file system - if (VFS_ROOT(vp->v_mount, &root_vp, ctx) != 0 || vp != root_vp) { - error = EINVAL; - vnode_put(vp); - goto outdrop; - } - vnode_put(root_vp); - - // get the owner of the root dir - VATTR_INIT(&va); - VATTR_WANTED(&va, va_uid); - if (vnode_getattr(vp, &va, ctx) != 0) { - error = EINVAL; - vnode_put(vp); - goto outdrop; - } - - // and last, check that the caller is the super user or - // the owner of the mount point - cred = vfs_context_ucred(ctx); - if (!is_suser() && va.va_uid != kauth_cred_getuid(cred)) { - error = EACCES; - vnode_put(vp); - goto outdrop; + if (error == 0) { + error = copyout((caddr_t)pathbufp, argp, pathlen); + } + if (error == 0) { + /* + * If the copyout was successful, now check to ensure + * that this vnode is not a BACKINGSTORE vnode. mtmd + * wants the path regardless. + */ + if (backingstore) { + error = EBUSY; + } + } + } else + (void)vnode_put(vp); } - - // if all those checks pass then we can mark the dependency - vfs_markdependency(vp->v_mount); - error = 0; - - vnode_put(vp); - - break; - } - - case F_GETPROTECTIONCLASS: { - // stub to make the API work - printf("Reached F_GETPROTECTIONCLASS, returning without action\n"); - error = 0; - goto out; - } - - case F_SETPROTECTIONCLASS: { - // stub to make the API work - printf("Reached F_SETPROTECTIONCLASS, returning without action\n"); - error = 0; - goto out; + FREE(pathbufp, M_TEMP); + goto outdrop; } - default: /* * This is an fcntl() that we d not recognize at this level; @@ -1621,10 +2413,19 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) * effectively overload fcntl() to send ioctl()'s. */ if((uap->cmd & IOC_VOID) && (uap->cmd & IOC_INOUT)){ - error = EINVAL; + error = EINVAL; goto out; } + /* Catch any now-invalid fcntl() selectors */ + switch (uap->cmd) { + case F_MARKDEPENDENCY: + error = EINVAL; + goto out; + default: + break; + } + if (fp->f_type != DTYPE_VNODE) { error = EBADF; goto out; @@ -1676,6 +2477,11 @@ fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval) kfree(memp, size); goto outdrop; } + + /* Bzero the section beyond that which was needed */ + if (size <= sizeof(stkbuf)) { + bzero ( (((uint8_t*)data) + size), (sizeof(stkbuf) - size)); + } } else { /* int */ if (is64bit) { @@ -1730,6 +2536,7 @@ out: * Parameters: p Process performing the dup * old The fd to dup * new The fd to dup it to + * fd_flags Flags to augment the new fd * retval Pointer to the call return area * * Returns: 0 Success @@ -1744,10 +2551,11 @@ out: * * Notes: This function may drop and reacquire this lock; it is unsafe * for a caller to assume that other state protected by the lock - * has not been subsequently changes out from under it. + * has not been subsequently changed out from under it. */ int -finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval) +finishdup(proc_t p, + struct filedesc *fdp, int old, int new, int fd_flags, int32_t *retval) { struct fileproc *nfp; struct fileproc *ofp; @@ -1758,9 +2566,8 @@ finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval) #if DIAGNOSTIC proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED); #endif - if ((ofp = fdp->fd_ofiles[old]) == NULL || - (fdp->fd_ofileflags[old] & UF_RESERVED)) { + (fdp->fd_ofileflags[old] & UF_RESERVED)) { fdrelse(p, new); return (EBADF); } @@ -1777,8 +2584,7 @@ finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval) proc_fdunlock(p); - MALLOC_ZONE(nfp, struct fileproc *, sizeof(struct fileproc), M_FILEPROC, M_WAITOK); - /* Failure check follows proc_fdlock() due to handling requirements */ + nfp = fileproc_alloc_init(NULL); proc_fdlock(p); @@ -1788,21 +2594,18 @@ finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval) return (ENOMEM); } - bzero(nfp, sizeof(struct fileproc)); - - nfp->f_flags = 0; nfp->f_fglob = ofp->f_fglob; - nfp->f_iocount = 0; #if DIAGNOSTIC if (fdp->fd_ofiles[new] != 0) - panic("finishdup: overwriting fd_ofiles with new %d\n", new); + panic("finishdup: overwriting fd_ofiles with new %d", new); if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) - panic("finishdup: unreserved fileflags with new %d\n", new); + panic("finishdup: unreserved fileflags with new %d", new); #endif if (new > fdp->fd_lastfile) fdp->fd_lastfile = new; + *fdflags(p, new) |= fd_flags; procfdtbl_releasefd(p, new, nfp); *retval = new; return (0); @@ -1821,6 +2624,7 @@ finishdup(proc_t p, struct filedesc *fdp, int old, int new, int32_t *retval) * * Returns: 0 Success * fp_lookup:EBADF Bad file descriptor + * fp_guard_exception:??? Guarded file descriptor * close_internal:EBADF * close_internal:??? Anything returnable by a per-fileops * close function @@ -1838,7 +2642,7 @@ close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retv { struct fileproc *fp; int fd = uap->fd; - int error =0; + int error; AUDIT_SYSCLOSE(p, fd); @@ -1849,11 +2653,18 @@ close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retv return(error); } + if (FP_ISGUARDED(fp, GUARD_CLOSE)) { + error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE); + (void) fp_drop(p, fd, fp, 1); + proc_fdunlock(p); + return (error); + } + error = close_internal_locked(p, fd, fp, 0); proc_fdunlock(p); - return(error); + return (error); } @@ -1877,10 +2688,9 @@ close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retv * * Notes: This function may drop and reacquire this lock; it is unsafe * for a caller to assume that other state protected by the lock - * has not been subsequently changes out from under it, if the - * caller made the call with the lock held. + * has not been subsequently changed out from under it. */ -static int +int close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags) { struct filedesc *fdp = p->p_fd; @@ -1897,13 +2707,13 @@ close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags) if ((fp->f_flags & FP_CLOSING) == FP_CLOSING) { - panic("close_internal_locked: being called on already closing fd\n"); + panic("close_internal_locked: being called on already closing fd"); } #if DIAGNOSTIC if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) - panic("close_internal: unreserved fileflags with fd %d\n", fd); + panic("close_internal: unreserved fileflags with fd %d", fd); #endif fp->f_flags |= FP_CLOSING; @@ -1941,11 +2751,13 @@ close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags) if (fp->f_flags & FP_WAITEVENT) (void)waitevent_close(p, fp); - if ((fp->f_flags & FP_INCHRREAD) == 0) - fileproc_drain(p, fp); + fileproc_drain(p, fp); - if (resvfd == 0) + if (resvfd == 0) { _fdrelse(p, fd); + } else { + procfdtbl_reservefd(p, fd); + } error = closef_locked(fp, fp->f_fglob, p); if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE) @@ -1954,14 +2766,14 @@ close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags) proc_fdunlock(p); - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + fileproc_free(fp); proc_fdlock(p); #if DIAGNOSTIC if (resvfd != 0) { if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) - panic("close with reserved fd returns with freed fd:%d: proc: %x\n", fd, (unsigned int)p); + panic("close with reserved fd returns with freed fd:%d: proc: %p", fd, p); } #endif @@ -2018,7 +2830,6 @@ fstat1(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsec struct user32_stat64 user32_sb64; } dest; int error, my_size; - int funnel_state; file_type_t type; caddr_t data; kauth_filesec_t fsec; @@ -2073,9 +2884,7 @@ fstat1(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsec break; case DTYPE_KQUEUE: - funnel_state = thread_funnel_set(kernel_flock, TRUE); - error = kqueue_stat(fp, sbptr, isstat64, p); - thread_funnel_set(kernel_flock, funnel_state); + error = kqueue_stat((void *)data, sbptr, isstat64, p); break; default: @@ -2302,10 +3111,7 @@ fpathconf(proc_t p, struct fpathconf_args *uap, int32_t *retval) } goto out; - case DTYPE_PSXSHM: - case DTYPE_PSXSEM: - case DTYPE_KQUEUE: - case DTYPE_FSEVENTS: + default: error = EINVAL; goto out; @@ -2925,68 +3731,6 @@ fp_getfpipe(proc_t p, int fd, struct fileproc **resultfp, return (0); } -#if NETAT -#define DTYPE_ATALK -1 /* XXX This does not belong here */ - - -/* - * fp_getfatalk - * - * Description: Get fileproc and atalk pointer for a given fd from the - * per process open file table of the specified process - * and if successful, increment the f_iocount - * - * Parameters: p Process in which fd lives - * fd fd to get information for - * resultfp Pointer to result fileproc - * pointer area, or 0 if none - * resultatalk Pointer to result atalk - * pointer area, or 0 if none - * Returns: EBADF The file descriptor is invalid - * EBADF The file descriptor is not a socket - * 0 Success - * - * Implicit returns: - * *resultfp (modified) Fileproc pointer - * *resultatalk (modified) atalk pointer - * - * Notes: The second EBADF should probably be something else to make - * the error condition distinct. - * - * XXX This code is specific to AppleTalk protocol support, and - * XXX should be conditionally compiled - */ -int -fp_getfatalk(proc_t p, int fd, struct fileproc **resultfp, - struct atalk **resultatalk) -{ - struct filedesc *fdp = p->p_fd; - struct fileproc *fp; - - proc_fdlock_spin(p); - if (fd < 0 || fd >= fdp->fd_nfiles || - (fp = fdp->fd_ofiles[fd]) == NULL || - (fdp->fd_ofileflags[fd] & UF_RESERVED)) { - proc_fdunlock(p); - return (EBADF); - } - if (fp->f_type != (DTYPE_ATALK+1)) { - proc_fdunlock(p); - return(EBADF); - } - fp->f_iocount++; - - if (resultfp) - *resultfp = fp; - if (resultatalk) - *resultatalk = (struct atalk *)fp->f_data; - proc_fdunlock(p); - - return (0); -} - -#endif /* NETAT */ - /* * fp_lookup * @@ -3037,6 +3781,72 @@ fp_lookup(proc_t p, int fd, struct fileproc **resultfp, int locked) } +/* + * fp_tryswap + * + * Description: Swap the fileproc pointer for a given fd with a new + * fileproc pointer in the per-process open file table of + * the specified process. The fdlock must be held at entry. + * + * Parameters: p Process containing the fd + * fd The fd of interest + * nfp Pointer to the newfp + * + * Returns: 0 Success + * EBADF Bad file descriptor + * EINTR Interrupted + * EKEEPLOOKING f_iocount changed while lock was dropped. + */ +int +fp_tryswap(proc_t p, int fd, struct fileproc *nfp) +{ + struct fileproc *fp; + int error; + + proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED); + + if (0 != (error = fp_lookup(p, fd, &fp, 1))) + return (error); + /* + * At this point, our caller (change_guardedfd_np) has + * one f_iocount reference, and we just took another + * one to begin the replacement. + */ + if (fp->f_iocount < 2) { + panic("f_iocount too small %d", fp->f_iocount); + } else if (2 == fp->f_iocount) { + + /* Copy the contents of *fp, preserving the "type" of *nfp */ + + nfp->f_flags = (nfp->f_flags & FP_TYPEMASK) | + (fp->f_flags & ~FP_TYPEMASK); + nfp->f_iocount = fp->f_iocount; + nfp->f_fglob = fp->f_fglob; + nfp->f_waddr = fp->f_waddr; + + p->p_fd->fd_ofiles[fd] = nfp; + (void) fp_drop(p, fd, nfp, 1); + } else { + /* + * Wait for all other active references to evaporate. + */ + p->p_fpdrainwait = 1; + error = msleep(&p->p_fpdrainwait, &p->p_fdmlock, + PRIBIO | PCATCH, "tryswap fpdrain", NULL); + if (0 == error) { + /* + * Return an "internal" errno to trigger a full + * reevaluation of the change-guard attempt. + */ + error = EKEEPLOOKING; + printf("%s: lookup collision fd %d\n", __func__, fd); + } + (void) fp_drop(p, fd, fp, 1); + } + return (error); +} + + /* * fp_drop_written * @@ -3150,9 +3960,14 @@ fp_drop(proc_t p, int fd, struct fileproc *fp, int locked) } fp->f_iocount--; - if (p->p_fpdrainwait && fp->f_iocount == 0) { - p->p_fpdrainwait = 0; - needwakeup = 1; + if (fp->f_iocount == 0) { + if (fp->f_flags & FP_SELCONFLICT) + fp->f_flags &= ~FP_SELCONFLICT; + + if (p->p_fpdrainwait) { + p->p_fpdrainwait = 0; + needwakeup = 1; + } } if (!locked) proc_fdunlock(p); @@ -3188,7 +4003,7 @@ fp_drop(proc_t p, int fd, struct fileproc *fp, int locked) * * The fileproc referenced is not returned; because of this, care * must be taken to not drop the last reference (e.g. by closing - * the file). This is inhernely unsafe, since the reference may + * the file). This is inherently unsafe, since the reference may * not be recoverable from the vnode, if there is a subsequent * close that destroys the associate fileproc. The caller should * therefore retain their own reference on the fileproc so that @@ -3249,7 +4064,7 @@ file_vnode(int fd, struct vnode **vpp) * * The fileproc referenced is not returned; because of this, care * must be taken to not drop the last reference (e.g. by closing - * the file). This is inhernely unsafe, since the reference may + * the file). This is inherently unsafe, since the reference may * not be recoverable from the vnode, if there is a subsequent * close that destroys the associate fileproc. The caller should * therefore retain their own reference on the fileproc so that @@ -3314,7 +4129,7 @@ file_vnode_withvid(int fd, struct vnode **vpp, uint32_t * vidp) * * The fileproc referenced is not returned; because of this, care * must be taken to not drop the last reference (e.g. by closing - * the file). This is inhernely unsafe, since the reference may + * the file). This is inherently unsafe, since the reference may * not be recoverable from the socket, if there is a subsequent * close that destroys the associate fileproc. The caller should * therefore retain their own reference on the fileproc so that @@ -3445,9 +4260,14 @@ file_drop(int fd) } fp->f_iocount --; - if (p->p_fpdrainwait && fp->f_iocount == 0) { - p->p_fpdrainwait = 0; - needwakeup = 1; + if (fp->f_iocount == 0) { + if (fp->f_flags & FP_SELCONFLICT) + fp->f_flags &= ~FP_SELCONFLICT; + + if (p->p_fpdrainwait) { + p->p_fpdrainwait = 0; + needwakeup = 1; + } } proc_fdunlock(p); @@ -3457,6 +4277,9 @@ file_drop(int fd) } +static int falloc_withalloc_locked(proc_t, struct fileproc **, int *, + vfs_context_t, struct fileproc * (*)(void *), void *, int); + /* * falloc * @@ -3481,7 +4304,7 @@ file_drop(int fd) * *resultfd (modified) Returned fd * * Locks: This function takes and drops the proc_fdlock; if this lock - * is alread held, use falloc_locked() instead. + * is already held, use falloc_locked() instead. * * Notes: This function takes separate process and context arguments * solely to support kern_exec.c; otherwise, it would take @@ -3490,22 +4313,38 @@ file_drop(int fd) */ int falloc(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx) +{ + return (falloc_withalloc(p, resultfp, resultfd, ctx, + fileproc_alloc_init, NULL)); +} + +/* + * Like falloc, but including the fileproc allocator and create-args + */ +int +falloc_withalloc(proc_t p, struct fileproc **resultfp, int *resultfd, + vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *arg) { int error; proc_fdlock(p); - error = falloc_locked(p, resultfp, resultfd, ctx, 1); + error = falloc_withalloc_locked(p, + resultfp, resultfd, ctx, fp_zalloc, arg, 1); proc_fdunlock(p); - return(error); + return (error); } +/* + * "uninitialized" ops -- ensure fg->fg_ops->fo_type always exists + */ +static const struct fileops uninitops; /* * falloc_locked * * Create a new open file structure and allocate - * a file decriptor for the process that refers to it. + * a file descriptor for the process that refers to it. * * Returns: 0 Success * @@ -3547,7 +4386,16 @@ int falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx, int locked) { - struct fileproc *fp, *fq; + return (falloc_withalloc_locked(p, resultfp, resultfd, ctx, + fileproc_alloc_init, NULL, locked)); +} + +static int +falloc_withalloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, + vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *crarg, + int locked) +{ + struct fileproc *fp; struct fileglob *fg; int error, nfd; @@ -3581,7 +4429,7 @@ falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, */ proc_fdunlock(p); - MALLOC_ZONE(fp, struct fileproc *, sizeof(struct fileproc), M_FILEPROC, M_WAITOK); + fp = (*fp_zalloc)(crarg); if (fp == NULL) { if (locked) proc_fdlock(p); @@ -3589,17 +4437,17 @@ falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, } MALLOC_ZONE(fg, struct fileglob *, sizeof(struct fileglob), M_FILEGLOB, M_WAITOK); if (fg == NULL) { - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + fileproc_free(fp); if (locked) proc_fdlock(p); return (ENOMEM); } - bzero(fp, sizeof(struct fileproc)); bzero(fg, sizeof(struct fileglob)); lck_mtx_init(&fg->fg_lock, file_lck_grp, file_lck_attr); fp->f_iocount = 1; fg->fg_count = 1; + fg->fg_ops = &uninitops; fp->f_fglob = fg; #if CONFIG_MACF mac_file_label_init(fg); @@ -3615,16 +4463,7 @@ falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, mac_file_label_associate(fp->f_cred, fg); #endif - lck_mtx_lock_spin(file_flist_lock); - - nfiles++; - - if ( (fq = p->p_fd->fd_ofiles[0]) ) { - LIST_INSERT_AFTER(fq->f_fglob, fg, f_list); - } else { - LIST_INSERT_HEAD(&filehead, fg, f_list); - } - lck_mtx_unlock(file_flist_lock); + OSAddAtomic(1, &nfiles); p->p_fd->fd_ofiles[nfd] = fp; @@ -3655,10 +4494,12 @@ falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, void fg_free(struct fileglob *fg) { - lck_mtx_lock_spin(file_flist_lock); - LIST_REMOVE(fg, f_list); - nfiles--; - lck_mtx_unlock(file_flist_lock); + OSAddAtomic(-1, &nfiles); + + if (fg->fg_vn_data) { + fg_vn_data_free(fg->fg_vn_data); + fg->fg_vn_data = NULL; + } if (IS_VALID_CRED(fg->fg_cred)) { kauth_cred_unref(&fg->fg_cred); @@ -3679,6 +4520,10 @@ fg_free(struct fileglob *fg) * that are either marked as close-on-exec, or which were in the * process of being opened at the time of the execve * + * Also handles the case (via posix_spawn()) where -all- + * files except those marked with "inherit" as treated as + * close-on-exec. + * * Parameters: p Pointer to process calling * execve * @@ -3686,34 +4531,34 @@ fg_free(struct fileglob *fg) * * Locks: This function internally takes and drops proc_fdlock() * - * Notes: This function drops and retakes the kernel funnel; this is - * inherently unsafe, since another thread may have the - * proc_fdlock. - * - * XXX: We should likely reverse the lock and funnel drop/acquire - * order to avoid the small race window; it's also possible that - * if the program doing the exec has an outstanding listen socket - * and a network connection is completed asyncrhonously that we - * will end up with a "ghost" socket reference in the new process. - * - * This needs reworking to make it safe to remove the funnel from - * the execve and posix_spawn system calls. */ void -fdexec(proc_t p) +fdexec(proc_t p, short flags) { struct filedesc *fdp = p->p_fd; int i; - struct fileproc *fp; + boolean_t cloexec_default = (flags & POSIX_SPAWN_CLOEXEC_DEFAULT) != 0; proc_fdlock(p); - i = fdp->fd_lastfile; + for (i = fdp->fd_lastfile; i >= 0; i--) { - while (i >= 0) { + struct fileproc *fp = fdp->fd_ofiles[i]; + char *flagp = &fdp->fd_ofileflags[i]; + + if (fp && cloexec_default) { + /* + * Reverse the usual semantics of file descriptor + * inheritance - all of them should be closed + * except files marked explicitly as "inherit" and + * not marked close-on-exec. + */ + if ((*flagp & (UF_EXCLOSE|UF_INHERIT)) != UF_INHERIT) + *flagp |= UF_EXCLOSE; + *flagp &= ~UF_INHERIT; + } - fp = fdp->fd_ofiles[i]; if ( - ((fdp->fd_ofileflags[i] & (UF_RESERVED|UF_EXCLOSE)) == UF_EXCLOSE) + ((*flagp & (UF_RESERVED|UF_EXCLOSE)) == UF_EXCLOSE) #if CONFIG_MACF || (fp && mac_file_check_inherit(proc_ucred(p), fp->f_fglob)) #endif @@ -3725,10 +4570,21 @@ fdexec(proc_t p) fdp->fd_lastfile--; if (i < fdp->fd_freefile) fdp->fd_freefile = i; + + /* + * Wait for any third party viewers (e.g., lsof) + * to release their references to this fileproc. + */ + while (fp->f_iocount > 0) { + p->p_fpdrainwait = 1; + msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, + "fpdrain", NULL); + } + closef_locked(fp, fp->f_fglob, p); - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + + fileproc_free(fp); } - i--; } proc_fdunlock(p); } @@ -3764,7 +4620,7 @@ fdexec(proc_t p) * thread making the call, rather than from the process. * * In the case of a failure to obtain a reference, for most cases, - * the file entry will be silently droppped. There's an exception + * the file entry will be silently dropped. There's an exception * for the case of a chroot dir, since a failure to to obtain a * reference there would constitute an "escape" from the chroot * environment, which must not be allowed. In that case, we will @@ -3822,7 +4678,7 @@ fdcopy(proc_t p, vnode_t uth_cdir) * our reference from the parent also * since the vnode has gone DEAD making * it useless... by dropping it we'll - * be that much closer to recyling it + * be that much closer to recycling it */ vnode_rele(fdp->fd_cdir); fdp->fd_cdir = NULL; @@ -3852,7 +4708,6 @@ fdcopy(proc_t p, vnode_t uth_cdir) FREE_ZONE(newfdp, sizeof *newfdp, M_FILEDESC); return(NULL); } - newfdp->fd_refcnt = 1; /* * If the number of open files fits in the internal arrays @@ -3905,10 +4760,14 @@ fdcopy(proc_t p, vnode_t uth_cdir) */ if (newfdp->fd_knlistsize != -1) { fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile]; - for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) { + flags = &newfdp->fd_ofileflags[newfdp->fd_lastfile]; + for (i = newfdp->fd_lastfile; + i >= 0; i--, fpp--, flags--) { + if (*flags & UF_RESERVED) + continue; /* (removed below) */ if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) { *fpp = NULL; - newfdp->fd_ofileflags[i] = 0; + *flags = 0; if (i < newfdp->fd_freefile) newfdp->fd_freefile = i; } @@ -3924,8 +4783,13 @@ fdcopy(proc_t p, vnode_t uth_cdir) flags = newfdp->fd_ofileflags; for (i = newfdp->fd_lastfile + 1; --i >= 0; fpp++, flags++) - if ((ofp = *fpp) != NULL && !(*flags & UF_RESERVED)) { - MALLOC_ZONE(fp, struct fileproc *, sizeof(struct fileproc), M_FILEPROC, M_WAITOK); + if ((ofp = *fpp) != NULL && + 0 == (*flags & (UF_FORKCLOSE|UF_RESERVED))) { +#if DEBUG + if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE) + panic("complex fileproc"); +#endif + fp = fileproc_alloc_init(NULL); if (fp == NULL) { /* * XXX no room to copy, unable to @@ -3933,10 +4797,8 @@ fdcopy(proc_t p, vnode_t uth_cdir) */ *fpp = NULL; } else { - bzero(fp, sizeof(struct fileproc)); - fp->f_flags = ofp->f_flags; - //fp->f_iocount = ofp->f_iocount; - fp->f_iocount = 0; + fp->f_flags |= + (ofp->f_flags & ~FP_TYPEMASK); fp->f_fglob = ofp->f_fglob; (void)fg_ref(fp); *fpp = fp; @@ -3977,28 +4839,23 @@ fdfree(proc_t p) proc_fdlock(p); - /* Certain daemons might not have file descriptors */ - fdp = p->p_fd; - - if ((fdp == NULL) || (--fdp->fd_refcnt > 0)) { + if (p == kernproc || NULL == (fdp = p->p_fd)) { proc_fdunlock(p); return; } - if (fdp->fd_refcnt == 0xffff) - panic("fdfree: bad fd_refcnt"); - /* Last reference: the structure can't change out from under us */ + extern struct filedesc filedesc0; + + if (&filedesc0 == fdp) + panic("filedesc0"); if (fdp->fd_nfiles > 0 && fdp->fd_ofiles) { for (i = fdp->fd_lastfile; i >= 0; i--) { if ((fp = fdp->fd_ofiles[i]) != NULL) { if (fdp->fd_ofileflags[i] & UF_RESERVED) - panic("fdfree: found fp with UF_RESERVED\n"); + panic("fdfree: found fp with UF_RESERVED"); - /* closef drops the iocount ... */ - if ((fp->f_flags & FP_INCHRREAD) != 0) - fp->f_iocount++; procfdtbl_reservefd(p, i); if (i < fdp->fd_knlistsize) @@ -4006,7 +4863,7 @@ fdfree(proc_t p) if (fp->f_flags & FP_WAITEVENT) (void)waitevent_close(p, fp); (void) closef_locked(fp, fp->f_fglob, p); - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + fileproc_free(fp); } } FREE_ZONE(fdp->fd_ofiles, fdp->fd_nfiles * OFILESIZE, M_OFILETABL); @@ -4033,55 +4890,6 @@ fdfree(proc_t p) FREE_ZONE(fdp, sizeof(*fdp), M_FILEDESC); } - -/* - * closef_finish - * - * Description: Called on last open instance for a fileglob for a file being - * closed. - * - * Parameters: fp Pointer to fileproc for fd - * fg Pointer to fileglob for fd - * p Pointer to proc structure - * - * Returns: 0 Success - * :??? Anything returnable by a per-fileops - * close function - * - * Note: fp can only be non-NULL if p is also non-NULL. If p is NULL, - * then fg must eith be locked (FHASLOCK) or must not have a - * type of DTYPE_VNODE. - * - * On return, the fg is freed. - * - * This function may block draining output to a character - * device on last close of that device. - */ -static int -closef_finish(struct fileproc *fp, struct fileglob *fg, proc_t p, vfs_context_t ctx) -{ - int error; - - - /* fg_ops completed initialization? */ - if (fg->fg_ops) - error = fo_close(fg, ctx); - else - error = 0; - - /* if fp is non-NULL, drain it out */ - if (((fp != (struct fileproc *)0) && ((fp->f_flags & FP_INCHRREAD) != 0))) { - proc_fdlock_spin(p); - if ( ((fp->f_flags & FP_INCHRREAD) != 0) ) { - fileproc_drain(p, fp); - } - proc_fdunlock(p); - } - fg_free(fg); - - return (error); -} - /* * closef_locked * @@ -4129,7 +4937,8 @@ closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p) * If the descriptor was in a message, POSIX-style locks * aren't passed with the descriptor. */ - if (p && (p->p_ladvflag & P_LADVLOCK) && fg->fg_type == DTYPE_VNODE) { + if (p && (p->p_ladvflag & P_LADVLOCK) && + DTYPE_VNODE == FILEGLOB_DTYPE(fg)) { proc_fdunlock(p); lf.l_whence = SEEK_SET; @@ -4139,7 +4948,7 @@ closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p) vp = (struct vnode *)fg->fg_data; if ( (error = vnode_getwithref(vp)) == 0 ) { - (void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context); + (void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context, NULL); (void)vnode_put(vp); } proc_fdlock(p); @@ -4164,7 +4973,13 @@ closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p) if (p) proc_fdunlock(p); - error = closef_finish(fp, fg, p, &context); + + /* Since we ensure that fg->fg_ops is always initialized, + * it is safe to invoke fo_close on the fg */ + error = fo_close(fg, &context); + + fg_free(fg); + if (p) proc_fdlock(p); @@ -4186,7 +5001,7 @@ closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p) * Locks: Assumes the caller holds the proc_fdlock * * Notes: For character devices, this occurs on the last close of the - * device; for all other file descriptos, this occurs on each + * device; for all other file descriptors, this occurs on each * close to prevent fd's from being closed out from under * operations currently in progress and blocked * @@ -4210,14 +5025,25 @@ fileproc_drain(proc_t p, struct fileproc * fp) if (fp->f_fglob->fg_ops->fo_drain) { (*fp->f_fglob->fg_ops->fo_drain)(fp, &context); } - if (((fp->f_flags & FP_INSELECT)== FP_INSELECT)) { - wait_queue_wakeup_all((wait_queue_t)fp->f_waddr, NULL, THREAD_INTERRUPTED); + if ((fp->f_flags & FP_INSELECT) == FP_INSELECT) { + if (wait_queue_wakeup_all((wait_queue_t)fp->f_waddr, NULL, THREAD_INTERRUPTED) == KERN_INVALID_ARGUMENT) + panic("bad wait queue for wait_queue_wakeup_all %p", fp->f_waddr); } + if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) { + if (wait_queue_wakeup_all(&select_conflict_queue, NULL, THREAD_INTERRUPTED) == KERN_INVALID_ARGUMENT) + panic("bad select_conflict_queue"); + } p->p_fpdrainwait = 1; msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, "fpdrain", NULL); } +#if DIAGNOSTIC + if ((fp->f_flags & FP_INSELECT) != 0) + panic("FP_INSELECT set on drained fp"); +#endif + if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) + fp->f_flags &= ~FP_SELCONFLICT; } @@ -4245,7 +5071,7 @@ fp_free(proc_t p, int fd, struct fileproc * fp) proc_fdunlock(p); fg_free(fp->f_fglob); - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + fileproc_free(fp); return(0); } @@ -4299,7 +5125,7 @@ flock(proc_t p, struct flock_args *uap, __unused int32_t *retval) if (how & LOCK_UN) { lf.l_type = F_UNLCK; fp->f_flag &= ~FHASLOCK; - error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx); + error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL); goto out; } if (how & LOCK_EX) @@ -4317,10 +5143,10 @@ flock(proc_t p, struct flock_args *uap, __unused int32_t *retval) #endif fp->f_flag |= FHASLOCK; if (how & LOCK_NB) { - error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK, ctx); + error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK, ctx, NULL); goto out; } - error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK|F_WAIT, ctx); + error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK|F_WAIT, ctx, NULL); out: (void)vnode_put(vp); out1: @@ -4329,7 +5155,6 @@ out1: } -#if CONFIG_EMBEDDED /* * fileport_makeport * @@ -4370,6 +5195,13 @@ fileport_makeport(proc_t p, struct fileport_makeport_args *uap, goto out; } + if (FP_ISGUARDED(fp, GUARD_FILEPORT)) { + proc_fdlock(p); + err = fp_guard_exception(p, fd, fp, kGUARD_EXC_FILEPORT); + proc_fdunlock(p); + goto out; + } + /* Dropped when port is deallocated */ fg = fp->f_fglob; fg_ref(fp); @@ -4465,15 +5297,13 @@ fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval) err = EINVAL; goto out; } - - MALLOC_ZONE(fp, struct fileproc *, sizeof(*fp), M_FILEPROC, M_WAITOK); + + fp = fileproc_alloc_init(NULL); if (fp == FILEPROC_NULL) { err = ENOMEM; goto out; } - bzero(fp, sizeof(*fp)); - fp->f_fglob = fg; fg_ref(fp); @@ -4483,6 +5313,7 @@ fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval) proc_fdunlock(p); goto out; } + *fdflags(p, fd) |= UF_EXCLOSE; procfdtbl_releasefd(p, fd, fp); proc_fdunlock(p); @@ -4491,7 +5322,7 @@ fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval) err = 0; out: if ((fp != NULL) && (0 != err)) { - FREE_ZONE(fp, sizeof(*fp), M_FILEPROC); + fileproc_free(fp); } if (IPC_PORT_NULL != port) { @@ -4500,7 +5331,6 @@ out: return err; } -#endif /* CONFIG_EMBEDDED */ /* @@ -4524,7 +5354,7 @@ out: * Notes: XXX This is not thread safe; see fdopen() above */ int -dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) +dupfdopen(struct filedesc *fdp, int indx, int dfd, int flags, int error) { struct fileproc *wfp; struct fileproc *fp; @@ -4571,11 +5401,18 @@ dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) */ switch (error) { case ENODEV: + if (FP_ISGUARDED(wfp, GUARD_DUP)) { + int err = fp_guard_exception(p, + dfd, wfp, kGUARD_EXC_DUP); + proc_fdunlock(p); + return (err); + } + /* * Check that the mode the file is being opened for is a * subset of the mode of the existing descriptor. */ - if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { + if (((flags & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { proc_fdunlock(p); return (EACCES); } @@ -4587,7 +5424,8 @@ dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error) fg_free(fp->f_fglob); fp->f_fglob = wfp->f_fglob; - fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; + fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd] | + (flags & O_CLOEXEC) ? UF_EXCLOSE : 0; proc_fdunlock(p); return (0); @@ -4623,10 +5461,11 @@ fg_ref(struct fileproc * fp) #if DIAGNOSTIC if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0) - panic("fg_ref: invalid bits on fp%x\n", (unsigned int)fp); + panic("fg_ref: invalid bits on fp %p", fp); if (fg->fg_count == 0) - panic("fg_ref: adding fgcount to zeroed fg :fp %x, fg%x\n ", (unsigned int)fp, (unsigned int)fg); + panic("fg_ref: adding fgcount to zeroed fg: fp %p fg %p", + fp, fg); #endif fg->fg_count++; lck_mtx_unlock(&fg->fg_lock); @@ -4656,7 +5495,7 @@ fg_drop(struct fileproc * fp) lck_mtx_unlock(&fg->fg_lock); } - +#if SOCKETS /* * fg_insertuipc * @@ -4751,7 +5590,7 @@ fg_removeuipc(struct fileglob * fg) lck_mtx_unlock(&fg->fg_lock); } } - +#endif /* SOCKETS */ /* * fo_read @@ -4907,3 +5746,31 @@ filetype_issendable(file_type_t fdtype) return FALSE; } } + + +struct fileproc * +fileproc_alloc_init(__unused void *arg) +{ + struct fileproc *fp; + + MALLOC_ZONE(fp, struct fileproc *, sizeof (*fp), M_FILEPROC, M_WAITOK); + if (fp) + bzero(fp, sizeof (*fp)); + + return (fp); +} + +void +fileproc_free(struct fileproc *fp) +{ + switch (FILEPROC_TYPE(fp)) { + case FTYPE_SIMPLE: + FREE_ZONE(fp, sizeof (*fp), M_FILEPROC); + break; + case FTYPE_GUARDED: + guarded_fileproc_free(fp); + break; + default: + panic("%s: corrupt fp %p flags %x", __func__, fp, fp->f_flags); + } +}